llama.cpp

context: ignore zero scale LoRAs when checking sameness

#20166

Merged